All posts by Simon

HP 3326A Two-Channel Synthesizer: replaced, and replaced again!

Recently, two assemblies of a non-working HP 3326A were fixed by replacing their 15 uF tantalum caps – a good number of them had failed, presumably, because of a bad production lot of these capacitors (see earlier post)

Unfortunately, during the test run, some sporadic failures of the power supply, with overcurrent indication flashing. Then, permanent failure of the -15 V rail – as it turns out, by a short in the assemblies we had just fixed! An again, a discolored tantalum capacitor. Replaced it, and a few hours later, the same issue, with another capacitor of the same kind.
My mistake, I had use a bag of cheap China-sourced 15 uF, 25 Volts dipped tantalums, but these seem to be no good (unlike other Chinese electronics good that have attained good quality in recent years, provided you don’t by the cheapest kind). Maybe it was my mistake to buy the cheapest tantalum capacitors, but not much choice if you need 34 pieces to fix some old equipment – I don’t want to pay EUR 1.45 each from top brand parts from Mouser or similar sources.

With some luck, I found reasonable prices KEMET T350 series Ultradip II capacitors, these are known to be reliable.

You can see the size differences – the KEMET part is much bigger than the Chinese 25 Volts part – it is more similar in size to a 15 Volts KEMET part. Probably, the design was put a bit to the limit.

With the capacitors replaced and another 48 hours of run in test – no issues at all and the 3326A can be considered fixed and working for now.

HP 4191A Impedance Analyzer: power supply fixed

In a recent post,HP 4191A, I have introduced a 4191A with defective power supply. In the meantime, a spare power supply regulator assembly arrived, and the repairs of the originally fitted, defective supply have been completed.

First, some research into the HP p/n 1826-0043 Opamp, which is considered to be a LM307 comparable part. But what is acutally inside the can? Let’s crack it open.

The die has a small marking, reading “LM107C”. The LM107 is very similar to the LM307, except for a wider temperature range, and somewhat extended power supply limits. In the current application, we can safely drop in LM307N opamps. It is also clearly visible that there are no bonds for external frequency compensation (like for the LM301).

The replacement power transistors have been discussed in the earlier post, and now all has been cleaned and new heat conducting grease added.

All the replaced parts marked, after all, I should have also replaced the fuses and fuse holds which show signs of corrosion and bad contact.

A quick test, on the 4191A, left hand side is the working spare, the right hand side, the repaired original supply. I didn’t connect it to the 4191A board, but just to the transformer to confirm the basic working condition without putting the 4191A to any dangers – the repaired supply with but just a spare, and probably it will never be needed… at least not for this unit.

There were no other defects with the 4191A, and now some tests with the 4191A working instrument. It is essentially a very precise one-port vector network analyzer, with a thermostated reflection bridge and some other features that make it suitable for component measurements. There are some (expensive) HP test fixtures, but essentially, you don’t need these because most of the components will need to be tested on some custom boards, or directly soldered to a test connector, to avoid parasitic effects at the frequencies of interest. I just used a set of SMA flange-type connectors.

First some standards were fabricated – a short, by applying a generous amount of solder the connector to completely short it, an open, by cutting off the center pin and machining it flat (maybe need to add a cap or other structure later, to avoid any parasitic capacities, for now, during calibration, you just need to keep the hand away (and any other ground planes). The load 50 Ohms – from two 100 Ohms SMS resistors in parallel.

Some test objects, a 680 pF SMD 0805 capacitor (NP0), a 100 Ohm SMD 0805 resistor, and a wired 100 Ohm metal film transistor.

The test itself, run with some excellent Excel software (by a certain Harry Percival) and Zplots (by Dan Maguire, AC6LA) via GPIB bus.

The 100 Ohm SMD resistor, it has pretty good performance out to 1 GHz.

The wired transistor, inductance is adding to the parasitic behavior.

… same in linear frequency scale.

The 680 pF capacitor.

Also did some drift checks, over 10 hours (after about 2 hours of warm up), I could not find any detectable drift of the Z values, so the instrument seems very stable at least with reference to the stability of the 50 ohms load measurement.

HP 8754A 4 MHz to 1300 MHz Network Analyzer: final repairs, and a function test

Finally, the spare parts arrived, and the repairs of the HP 8754A could be finalized. The LM339 comparators, fitted to the boards…

The cap of the mains filter had many small cracks – replaced. For some reasons, the original filter had a Y-rate cap across the mains supply – Y rating is usually for connection from mains to earth. So I replaced it with a X2 rated cap for service parallel to mains.

Some tests – the 8754a is a very nice unit, because of its instantaneous response to the dial settings, rather than the delay of any digital network analyzer. Even the most modern of all units still don’t such a direct feel compared to the fully-analog 8754a.

HP/Agilent 6060A System DC Electronic Load: a quick repair

This is a 300 Watts, 60 Volts, 60 Amp electronic load, a quite handy device to have, especially, a HP/Agilent brand item. There are many cheap electronic loads, but I would rather recommend to get a good instrument, if you want to put some power supplies to real tests. Otherwise, you load may fail earlier than the supply.

The instrument we are dealing with here, a low cost auction fid – it had a bad front connector. These instrument use HP 60 Amp binding posts, these are quite rare and expensive (about EUR 40 per piece from Keysight), and the plastic gets brittle over time, and with overtightening it can break. The instrument had front and rear connections, I only need one set – so it will be an easy repair by just moving the good binding posts to the front.

Also, we find that all the X and Y rated capacitors have hair cracks, and are of RIFA brand, so these may fail soon – let’s replace them all.

The power is dissipated in several MOSFETs, all mounted to a large heatsink. Essentially, a small 300 Watts room heater, which is great to have these days in cold Japan.

The front connectors, after repair (just moved the rear connectors to the front, rear connectors, I don’t need them).

New caps soldered in – quite a difficult task because some vias are part of large copper fills, without thermal relieve, and I don’t want to preheat the whole board.

Finally managed to solder-in the X and Y capacitors.

A test at 40 Volts, 6 Amps, running for several hours with no issues at all!

Agilent 4352B VCO/PLL Signal Analyzer: working!

After a short xmas vacation, several spare parts arrived, including, 10 amp solder-in fuses, and thermal glue (704 silicon glue).

The glue is needed to mount the defective/blown thermal fuse to the power resistor. This resistor usually stays cool but will heat up in case of a power supply failure.

The fuse protects the primary of the switchmode transformer, it is a 10 Amp fuse, and it took a while to find it – it is located in a hidden place underneath the transformer.

Now, with the fuse installed, the thermal fuse glued to the resistor, and the two drive mostfets replaced, the Artsyn 24 Volts supply is starting up just fine. All self-tests passed!

Next step, let’s update the firmware, and do some tests.

The firmware version 2.11 is the latest one available, but it needs to be loaded from a 3.5 inch floppy – I have a USB floppy drive here, and one single disc which I purchased from Sri Lanka. Took a few attempts to convince the 4352B to read the disc and load the firmware. But finally, success!

Many tests could be done, here just a simple test with a 15 MHz signal from a 3585A vs. a 8642B generator. Seems to work well, and easy to use.

Now we can close the case, and use the device for VCO characterization, phase noise measurement, etc.

HP 6038A System Power Supply: all fixed!

After some weeks, the spare parts arrived – RIFA X2/Y2 rated capacitors (now made by Kemet), a full set (see earlier post, 6038a repair).

The new X2 capacitor, let’s hope RIFA has improved the resin and durability. Albeit, the old capacitors lasted for a long time…

And a fan, from China. The fan, upon close inspection, it has a broken frame, but fair enough, I will use this one while a replacement is on the way.

A lot of dust removed from the case and boards, all completely disassembled. The X and Y capacitors all replaced – the old capacitors are still working, but cracked and it is good practice to replace them, unless, you want to risk a lot of smoke and stench (usually, at least no fire risk).

Always good to use high quality tools – I only have low quality tools here, and bits that crack!

All cleaned and put together…

…finally, some testing. It is working, the fan is providing a substantial amount of cooling, it is definitely big enough for the unit.

HP 3326 Two-Channel Synthesizer: a bad lot of tantalum capacitors

This HP 3326A was found for a ridiculously low price, non-working, so I decided to pick it up, in case I need some spares for my good 3326A, or as a source for some HP parts. But when it arrived, it was in such good shape that a repair appeared worthwhile.

The symptom, it just doesn’t start up, the +15 Volt and -15 V rails shorted. Brief check showed that the power supply is working. There must be a short somewhere in one of the modules. How to find such short?

The 3326A has a cast aluminum cage construction, which houses all the modules in separate cavities, all heavy cast metal! To find the defective module, we first have to undo 100s of screws… and usually the last module will be the one at fault.

Half an hour later – found that both (!) phase detector boards – these are identical for channel A and B – have shorts.

Some more probing later – the reason a couple of shorted tantalum caps, 15 microfarads.

These are in general high quality capacitors, but it must have been a bad batch. So, let’s desolder all the 15 uF caps, and solder in new ones.

Even with the dead caps cut-out, the 3326A is working again! No issues with any of the self tests, including the service self test (push button self test-%-6 to activate).

Some maintenance is also needed on the power supply. Checked all the transistors which are known to develop issues with the sockets. And added some thermal compound to the 5 Volt rail transistor which was running a bit hot.

Also, the connector to the transformer has the typical bad soldering and cracked solder joints, all now re-soldered with a generous amount of good old lead containing solder.

This unit even has option 001, the precision ovenized reference.

There is no manufacturer datasheet available for this Japanese OCXO, but the HP manual has all the data. It is not an ultra-stable timing standard, but by far good enough for a two tone synthesizer.

HP 8754A Network Analyzer: gold, sapphire and still low output

Some performance validation of the recently fixed 8754A revealed that the output is leveled at 0 dBm, but it doesn’t provide any more than 2 dBm, when you turn the knob to higher levels… it should provide at least +10 dBm leveled, and +13 dBm typical.

So, what is wrong? The signal source is mostly located on the A7 assembly, two VCOs, a mixer, and several amplifiers and levelers.

The osciallator is working, as we can see, but the amplifier circuit, a golden box, number 5086-7235, is not amplifying sufficiently. HP did not consider this field-repairable, so the manual only has some rough information about its contents.

To find out more, we need to crack it open – it is not welded, but glued with a generous amount of silver epoxy.

From right to left, the preamplifier, a filter (LC low pass to remove the VCOs and higher mixer products), and the power amplifier with detector and a -20 dB tap for the PLL-marker circuit.

I checked all the bias voltages and currents, these seem OK. The main amp substrate (sapphire?) has a crack, but is it not fully going through the material, and the gold layer is thick, and the crack is not cutting through the critical sections.

The filter inductances, gold traces on alumina, with some bonding wire. I assume, hand made… It is a pitty I don’t have a microfabrication facility at my disposal, and a wire bonding machine…

Various testing has been performed, to find out the power levels, using a 50 MHz precision source, and a fine tipped probe to check the levels on the substrate (using a microscope, and a steady hand to avoid damaging the bond wires).

From the data it is obvious that the preamp is not amplifying, but absorbing power. This is good and bad, because the final amp needs to provide clean amplification to avoid spurious, so I don’t want to mess with it, and there is also the detector diode, which is essential for the flatness of the unit, also something that is not easily fixed, if you introduce some parasitic resonances or the like.

The fix – scraped off the transistors and most of the gold from the preamp, and soldered a short wire across the substrate, I think it is about 50 ohms impedance. Then, I inserted a set of 2 integrated microwave amplifiers (a MSA-0505 and MSA-0386) to provide about 19 dB gain.

The maximum output +13 dBm at virtually all frequencies (a small dip around 1 GHz).

A test at various power levels, with a good spectrum analyzer (don’t have a calibrated power meter here, but this analyzer is pretty well calibrated). Amplitude is 1 dB per division. The 8754A is calibrated at 0 dBm and 10 dBm, at 50 MHz.

0 dBm leveled output, 1 to 1400 MHz… pretty good.

10 dBm output, also, great flatness.

Finally, at test at 5 dBm – it’s accurate and flat!

Now, we will let this run for several hours at maximum output, to see if the repair is permanent, then the amps will be sealed with epoxy (just plain epoxy, no silver epoxy).

HP 5086-7803 YIG tuned Filter and Switch: “SYTF” details and adjustment

With the repair of many spectrum analyzers, it turns out, the preselectors are usually not easily damaged, because of their self limiting characteristics, and because of the absence of active parts. The SYTF is diffient in that it has an active part, a switch.

It is already some years old, but no reason why such parts should have much aging at all.

The symptom of this unit, it is working in the low frequency region setting, LOW band out, but the high band is dead, loss in excess of 35 dB, at all frequencies, and independent of the tuning current supplied.

So my first assessment was, this unit needs replacing, and I found a replacement part online, from a US seller, not cheap, but OK, the 8561E analyzer is worth it, if it is working again after the repair. Unfortunately, several week of waiting were all for nothing – the seller shipped the wrong part and it took a while to get the money back, but finally all settled, except, we still have the defective filter.

Let’s try to investigate the nature of the defect, and open it up. Fortunately, these filters are not hermetically welded like some other YIG parts.

You can clearly see the coil, the inlet and outlets (low and high band) by rigid SMA cable (1 mm size!).

First, let’s study the switch. It is not actually switching the high band, as I originally assumed, but it is switching on and off the low band.

It is a series-shunt-series type FET switch, controlled by about -10 V negative voltage (1 kOHM vs. -15 V connection is the usual control method, floating or ground to switch off).

I could not find the exact die and model for this switch, but there are many similar models that clearly show the structure. The shunt and double series construction will provide very high isolation.

After removing carefully the gold mesh (it is only lightly glued on, I will use some tiny traces of epoxy to stick it back on), some study under the microscope.

Clearly, the spheres are misaligned! The spheres must be placed in the center of the coupling loop, to allow for RF to couple. Generally speaking, during alignment, the sphere is only turned, and then the position fixed by some epoxy – which all seems to be intact, and solidly fixed. So what has happended? I think it has to do with the mounting blocks, which are of different material compared to the based (which needs to be magnetic Fe-Nickel alloy). With frequent temperature cycling, I believe there is some migration of the mounting blocks, fractions of a micrometer every time (keep in mind, the YIG spheres are heated during operation). While we can speculate about the reason of the migration, the result is clear, and the action as well: we need to realign the spheres.

I decided not to undo the screws because the coupling loops can be easily damaged, and used a screwdriver to carefully push the mounting blocks away from the coupling section, bit by bit, under control with a microscope.

Finally managed to get all the sphere properly aligned. If you don’t know how it works, never turn the YIG spheres! These need to be aligned for thermal stability effects, not only amplitude – something which you may have trouble doing at home.

After all the alignment, a quick test setup, with a current source for the main coil, and another supply for the heater and switch connection. Note that the current source is set in parallel with a capacitor (22 uF) to allow for stable regulation with the strongly inductive load.

The insertion loss test – done by checking at several frequencies, using the lines of a good comb generator.

The insertion loss, in my simple setup, it is about 4-6 dB for a 3 stage filter, not bad. I don’t know the original performance spec, but it is definitely in the typical range of such filters, and good enough for a spectrum analyzer in any case (in the worst case, we will loose 1-2 dB of sensitivity). Maybe I will eventually find a new filter at a reasonable price, to check it – it could also have moved spheres.

The tuning current is very linear, I don’t expect any issues with using this part in the analyzer (the tuning characteristics can be programmed and stored in the EEPROM of the analyzer, to control its DAC appropriately, also, we will need to recalibrate the flatness).

HP 8754A 4 MHz to 1300 MHz Network Analyzer: an analog computer, and a few rusted transistors

There is no specific need for a 1.3 GHz Network Analyzer in my workshop, because there are already several more modern instruments, but this HP 8754A is a real marvel, it was original designed as a “moderately priced, compact” type network analyzer, whatever was considered moderate by HP at the time (maybe the value of two or three small cars?). Finding the offer for a rediculouly low price, for a non-working unit, on Yahoo Japan, I could not resist to place a very moderate bid. Turns out, I was to only bidder, in whole Japan. My original thought was to use it for some experiments, and then, use it a as a source of HP spare parts (there are many FETs, Opamps, transitors, etc. in this machine).

Once the unit arrived, I powered it up, only to find out two things – the -10 Volt and +5 Volt power supplies are not working. And the CRT is very good and sharp. Maybe not many hours of use. Also, the unit is generally clean and in original condition – no other repair attemped. Even the HP instrument feet were included.

The -10 Volt, it required some troubleshooting of the low voltage assembly (corroded transistor legs), see below. The +5 supply, the issue could be traced to a defective TO-3 HP 1820-0430 integrated regulator, alias LM309K.

This regulator is mounted on an aluminum plate in the chassis, with some rather thick ceramic insulator, and what appeared to be only traced of thermal grease. Usually, these are protected against short and overheating, maybe, it was just running a bit hot for year, eventually, accelerating aging and finally triggering natural/random failure with no external even. We will never know, we only know, we have to fix it.

The LM309K tends to become rare and expensive, I still have some back at the Ludwigshafen, Germany workshop, but not here in my temporary Japanese workshop. Checking the offers, I found some very inexpensive LM323K.

The LM323K, it is a very similar device – just higher current capability. It is not critical for the 8754A, the 5 V rail is only loaded by about 0.25-0.3 Amp (as checked with a power supply).

Now, to the corroded transistors. This only seems to affect the boards thats are close to the air inlet, maybe some contamination from ambient air (salt?) is accelerating the effect, related to gold plated steel wire leg transistor. Other transistors have copper, or special alloy wire, but especially the “4-404” and 2N2222A transistors used by HP in the late 1970s seems to be affected by this phenomenon. Not so much in dry countries, but in instruments subject to humind and salty (sea?) conditions here in Japan – just a few km from the cost in most cases.

The 4-404 transistor, alias SS9333, 1854-0404 HP part number, it is a kind of mystery, no data available, and I have seen this part in may Hp instruments, always replacing them with some 4-404 scavenged from part units, etc. But for the 8754A, should we really buy some expensive old HP parts or wait for a long time to go back to Germany to the parts storage? Time for some characterization – found one good, only slightly rusted 4-404, and did some gain, DC performance and frequency response tests at typical currents.

Some basic data could be found – nothing special, the voltage rating rather moderate, and power rating, as well.

Some tests and calculations, it is medium to high gain NPN transistor. BC337-25 or BC337-40 can be valid replacments, I used BC337-25, selected for a gain of 250-300.

High frequency performance is nothing special, it can be easily met by a BC337.

Several 2N2222A are a bit easier to replace – just replaced the TO-18 metal can units with some generic TO-92 2N2222A (or whatever silicon fragments the mass producers put into the 2N2222A case nowadays).

After these initial fixes, the power is up, and the transistors all good. Initial assessment –
(1) front panel “analog computer” is working, some contact cleaner will do the trick, there is no mechanical damage
(2) the CRT will need a filter, it is missing.
(3) the RF output seems to work, at least there is power – need to check with a counter and properly align the linearity, etc.
(4) The VCO and PLL of the receivers seems to have some trouble, but the samplers are working! That’s a relieve.

See below this is a 35 MHz input signal, sampled with the VCO at about 33 MHz, giving a 1 MHz frequency.

So, what is wrong with the PLL? The PLL, it’s purpose is to have a line of a comb generator/multiplier (which is generating the sampler pulses by a step recovery diode) always 1 MHz away from the RF, to give a 1 MHz IF for the R, A and B channels.

This is achieved by first pretuning the VCO, by setting a frequency close to the needed multiple of the VCO, then the PLL is activated and phase lock achieved.

The phase detector, first, I thought is not working, because there is no proper output. But once desoldered, all the transistors tested OK.

The pretune, also this seems to be working, but hold – it is working too well! It is overruling the phase detector.

Further study shows that there are FET switched controlled by a logic signal, via a LM339 comparator. And, as it turns out, the LM339 is dead (both switches on)…

Temporarily fixed the issue by disabling the pre-tune, and enabling the PLL – and, it does lock (albeit, not a fast sweeps – which needs the pretune). But it works of you slowly increase the frequency starting from 0 MHz (this way, you can even measure as 1000 MHz, phase locked!).

After the PLL had been fixed, still some more issues – the R channel detector is not giving a proper output (switching the A8 and A11 boards showed, that the A8 board, which is the same board but used for the A-B channel is working!). Also some issues with the IF switching of the A/B channel, let’s fix this first. The IF switch is part of the A6 assembly mentioned before (which has the VCO and comb generator-diode pulser). To check it, without any fancy extender boards, you can just solder a few wires to the board. I generally prefer solid core telephone wire, this has a very strong and thin insulation, and doesn’t cause shorts easily, because of the single, solid core.

Also here, a dead LM339! Hardwired it for now to conduct the A channel IF.Ordered some LM339N, 10 pcs for USD 1.37.

Now, a few general views, top view:

You can see the card cages, power supply, and the RF sections with oscillators, mixers, samplers.

The bottom side, there are several dangerous DC voltages exposed, don’t touch!

The remaining issue, fixing the R detector and log amplifier, assembly A11. After some probing and thanks to having a working assembly (the A8 A/B detector assembly), the fault could be traced to the log amplifier, and furtunately, not to the transistor pair, which would be very difficult to source or replace, but to the reference amplifier, U2. This is a simple LM301, alias HP 1820-0223.

The LM301, a really early Signetics model! Unfortunately, it is dead, the inputs are somehow leaking negative current.

I already have some LM301 on order, but for the time being, used an old LM301, slightly rusted that I had desoldered recently from a 4191A power supply.

After all these fixed, the unit’s basic functions have all been restored. Sure, there will be through alignement and check, but I will do this once the LM339 and LM301 have made it to Japan. Checked for other issues, by running the units for several hours – very stable. To track the frequency stability, I used a 830 MHz bandpass filter.


Also, the instrument originally came with a plastic printed Smith chart that can be attached to the CRT. Wanted to print one, or have one made by photo printing on lightsetting film. But this is more for decorative purposes, and can be done later.